Matrix Completion for the Prediction of Yearly Country and Industry-Level CO2 Emissions

In the recent past, yearly CO2 emissions at the international level were studied from different points of view, due to their importance with respect to concerns about climate change. Nevertheless, related data (available at country-industry level and referred to the last two decades) often suffer from missingness and unreliability. To the best of our knowledge, the problem of solving the potential inaccuracy/missingness of such data related to certain countries has been overlooked. Thereby, with this work we contribute to the academic debate by analyzing yearly CO2 emissions data using Matrix Completion (MC), a Statistical Machine Learning (SML) technique whose main idea relies on the minimization of a suitable trade-off between the approximation error on a set of observed entries of a matrix (training set) and a proxy of the rank of the reconstructed matrix, e.g., its nuclear norm. In the work, we apply MC to the prediction of (artificially) missing entries of a country-specific matrix whose elements derive (after a suitable pre-processing at the industry level) from yearly CO2 emission levels related to different industries. The results show a better performance of MC when compared to a simple baseline. Possible directions of future research are pointed out.